We support running the experiments via the provided Docker image.
If you want to install the repository manually, please refer to the recipe contained in the
Dockerfile
.
- Pull the docker image:
docker pull ghcr.io/hsp-iit/mask-ukf:latest
- Launch the container:
docker run -it --rm --user user --env="DISPLAY" --net=host ghcr.io/hsp-iit/mask-ukf:latest
- Clone and build the project:
git clone https://github.com/hsp-iit/mask-ukf.git cd mask-ukf mkdir build && cd build cmake ../ make
- Download and unzip the accompanying data
and the YCB-Video model set:
cd /home/user/mask-ukf bash misc/download_accompanying_data.sh bash misc/download_ycb_models.sh
- Run the experiments (optional):
cd /home/user/mask-ukf bash test/test_all.sh
The accompanying data contains the pre-evaluated results. If desired, the results can be re-evaluated using the above command.
- Run the evaluation:
cd /home/user/mask-ukf bash evaluation/evaluate_<mask_set>_<metric>_<algorithm>.sh
<mask_set>
can bemrcnn
(Mask R-CNN) orposecnn
(PoseCNN),<metric>
can beadd_s
(ADD-S) orrmse
(RMSE) and<algorithm>
can be empty (MaskUKF),icp
(ICP) ordensefusion
(DenseFusion). - Visualize the results:
cd /home/user/mask-ukf python3 evaluation/renderer/renderer.py --algorithm <algorithm> --mask_set <mask_set> --object <object_name> --video_id <video_id>
<algorithm>
can bemask-ukf
(MaskUKF),icp
(ICP) ordense_fusion
(DenseFusion),<mask_set>
is as above,<object_name>
is e.g.002_master_chef_can
and<video_id>
is the YCB-Video video id, e.g.0048
.
In order to run the visualizer it could be required to temporarily execute
xhost +
in a console outside of Docker in order to allow the container accessing the X server facilities. The command can be run even after the container has been already launched.
If you find the MaskUKF code useful, please consider citing the associated publication:
@ARTICLE{10.3389/frobt.2021.594583,
AUTHOR={Piga, Nicola A. and Bottarel, Fabrizio and Fantacci, Claudio and Vezzani, Giulia and Pattacini, Ugo and Natale, Lorenzo},
TITLE={MaskUKF: An Instance Segmentation Aided Unscented Kalman Filter for 6D Object Pose and Velocity Tracking},
JOURNAL={Frontiers in Robotics and AI},
VOLUME={8},
PAGES={38},
YEAR={2021},
URL={https://www.frontiersin.org/article/10.3389/frobt.2021.594583},
DOI={10.3389/frobt.2021.594583},
ISSN={2296-9144}
}
and/or the repository itself by pressing on the Cite this respository
button in the About section.
This repository is maintained by:
![]() |
@xenvre |